On Thu, 2008-03-20 at 21:35 +0100, Carsten Otte wrote: > Dave Hansen wrote: > > Well, and more fundamentally: do we really want dup_mm() able to be > > called from other code? > > > > Maybe we need a bit more detailed justification why fork() itself isn't > > good enough. It looks to me like they basically need an arch-specific > > argument to fork, telling the new process's page tables to take the > > fancy new bit. > > > > I'm really curious how this new stuff is going to get used. Are you > > basically replacing fork() when creating kvm guests? > No. The trick is, that we do need bigger page tables when running > guests: our page tables are usually 2k, but when running a guest > they're 4k to track both guest and host dirty&reference information. > This looks like this: > *----------* > *2k PTE's * > *----------* > *2k PGSTE * > *----------* > We don't want to waste precious memory for all page tables. We'd like > to have one kernel image that runs regular server workload _and_ > guests. That makes a lot of sense. Is that layout (the shadow and regular stacked together) specified in hardware somehow, or was it just chosen? What you've done with dup_mm() is probably the brute-force way that I would have done it had I just been trying to make a proof of concept or something. I'm worried that there are a bunch of corner cases that haven't been considered. What if someone else is poking around with ptrace or something similar and they bump the mm_users: + if (tsk->mm->context.pgstes) + return 0; + if (!tsk->mm || atomic_read(&tsk->mm->mm_users) > 1 || + tsk->mm != tsk->active_mm || tsk->mm->ioctx_list) + return -EINVAL; -------->HERE + tsk->mm->context.pgstes = 1; /* dirty little tricks .. */ + mm = dup_mm(tsk); It'll race, possibly fault in some other pages, and those faults will be lost during the dup_mm(). I think you need to be able to lock out all of the users of access_process_vm() before you go and do this. You also need to make sure that anyone who has looked at task->mm doesn't go and get a reference to it and get confused later when it isn't the task->mm any more. > Therefore, we need to reallocate the page table after fork() > once we know that task is going to be a hypervisor. That's what this > code does: reallocate a bigger page table to accomondate the extra > information. The task needs to be single-threaded when calling for > extended page tables. > > Btw: at fork() time, we cannot tell whether or not the user's going to > be a hypervisor. Therefore we cannot do this in fork. Can you convert the page tables at a later time without doing a wholesale replacement of the mm? It should be a bit easier to keep people off the pagetables than keep their grubby mitts off the mm itself. -- Dave _______________________________________________ Virtualization mailing list Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx https://lists.linux-foundation.org/mailman/listinfo/virtualization
